AAlexTYJAdd streaming ASR inference
| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
add VibeVoice-Realtime | 8 个月前 | |
Add streaming ASR inference VibeVoice-ASR-Streaming transcribes while the audio is still arriving, emitting text once per audio chunk instead of waiting for the recording to end. Three pieces make that work on the transformers path: - streaming_generate / init_streaming_state / streaming_generate_step on the ASR model, so a caller can either hand over a whole tensor and iterate chunks, or drive the KV cache chunk by chunk as audio arrives. - text_chunk_end_id on the tokenizer, read from the vocabulary and left as None when absent, so non-streaming checkpoints are unaffected. - The processor now prefers a checkpoint's own tokenizer files, because <|text_chunk_end|> lives there and not in the base LM vocabulary; a language_model_pretrained_name that gets overridden is warned about rather than dropped silently. Adds a file-based inference script and an in-process FastAPI demo page, plus docs/vibevoice-asr-streaming.md. | 12 天前 | |
Add streaming ASR inference VibeVoice-ASR-Streaming transcribes while the audio is still arriving, emitting text once per audio chunk instead of waiting for the recording to end. Three pieces make that work on the transformers path: - streaming_generate / init_streaming_state / streaming_generate_step on the ASR model, so a caller can either hand over a whole tensor and iterate chunks, or drive the KV cache chunk by chunk as audio arrives. - text_chunk_end_id on the tokenizer, read from the vocabulary and left as None when absent, so non-streaming checkpoints are unaffected. - The processor now prefers a checkpoint's own tokenizer files, because <|text_chunk_end|> lives there and not in the base LM vocabulary; a language_model_pretrained_name that gets overridden is warned about rather than dropped silently. Adds a file-based inference script and an in-process FastAPI demo page, plus docs/vibevoice-asr-streaming.md. | 12 天前 | |
spelling: settable Signed-off-by: Josh Soref <2119212+jsoref@users.noreply.github.com> | 4 个月前 | |
fix: remove unused checkpoint conversion script with unsafe torch.load (CWE-502) The nnscaler checkpoint conversion script used torch.load() without weights_only=True, allowing arbitrary code execution via malicious .pt files on PyTorch < 2.6. The script is not referenced anywhere in the codebase and is not needed by end users, so removing it entirely eliminates the vulnerability. Reported-by: Offgrid Security (https://www.offgridsec.com) | 4 个月前 | |
feat: add __init__.py files to enable module imports Add __init__.py files to vibevoice/modular and vibevoice/processor directories to properly export classes and enable package imports. This allows users to import the package after installation: - from vibevoice import VibeVoiceStreamingForConditionalGenerationInference - from vibevoice.modular import VibeVoiceStreamingConfig - from vibevoice.processor import VibeVoiceStreamingProcessor Fixes import errors when using `pip install -e .` | 8 个月前 |